2000 AAAI Fall Symposium "Socially Intelligent Agents: The Human in the Loop"

Working Group on Emotions in Social Interactions

Notes by Lola Cañamero (LRI, University of Paris-XI, France)

Topics discussed:

1) What do SIA need emotions for?

• To convey intentionality
• To elicit emotion in others, to influence others
• To make humans feel more comfortable in the interactions with artifacts
• To better understand communication, at a "deeper" level: we want artifacts understand what we mean, not just what we say

2) Which emotions are better suited for SIA?

a) "Basic" emotions:

• Used in most emotional artifacts nowadays
• Typical subset: anger, happiness, sadness, fear, disgust, surprise
• They are easier to recognize (both by humans and artificially); also easier to synthesize
• They are adequate for simple interactions (e.g., with children), but can be a bit limited for more complex ones

b) More "subtle" emotions:

We should start investigating/using these, at several levels:

• Artifacts recognizing our subtle emotions. This raises privacy issues: to what extent do we want/is it ethical that artifacts recognize our most subtle emotions?
• Artifacts having/showing more subtle emotions. Possible drawback: an artifact with too complex emotional states might be much more demanding for the human user (difficult of the artifact's emotions becomes more difficult, user might need to care more about the artifact's "feelings", etc)

3) (How) will humans accept/trust emotional artifacts?

At least two factors are essential for this:

• Believability
• Human feels s/he is in control of the interaction at critical points.

What these two things mean depends on factors such as: personality of the user, application/domain, culture, context, etc.

Proposal: we should model emotions adapting them to types of "profiles" that take these factors into account. Also take into account the evolution of emotional interaction over time.

4) Coherence and consistency should be preserved:

• Between expressed emotion and contextual clues
• Among "channels" or sensory and expressive modalities.
• In the dynamics of the interaction; issues such as:
  • control of own's emotions in social interactions
  • "honest" versus "fake" emotional displays: can our technology detect this? Is it ethical to try to detect the emotion purposefully hidden? Relation to machiavellian use of emotions.

5) Some needs were identified, such as:

• Multi-modal interaction
• Detection and use of clues from context, personality, individual differences
• More subtle emotions for more complex interactions

6) Anthropomorphism:

• Can we / is it sensible to implement in artifacts the same methods that humans use, e.g. when perceiving, recognizing emotions, etc? What is the value of bio-sensors for emotion recognition?
• Potential dangers of extreme realism. A robot is a robot, not a human. We want humans to be able to connect with it, to feel comfortable in interactions, but not to fool them into believing they are interacting with another human.

As a conclusion: it can be very valuable to rely on the human tendency to anthropomorphize and to treat artifacts as social partners, but explicitly including anthropomorphism in design is less advisable.

Participants:

• Alan Bond
• Lola Canamero
• Cristina Conati
• Nadja De Carolis
• Jakob Fredslund
• Eva Hudlicka
• Christine Lisetti
• Valery Petrushin

Back to Program "Socially Intelligent Agents - The Human in the Loop"